Human Review
Review Manusia
The user reading the code the agent produced and forming a judgement on it. Reading the diff counts; reading the summary does not.
Human review is a person reading what the agent produced and forming a judgement about it. The critical detail is what counts as reading: reading the diff counts, reading the agent's summary of the diff does not.
That distinction is where the value leaks out. A fluent summary of a change is cheaper to produce than the change itself, so a reviewer who skims it is reviewing the model's account of the work rather than the work. The output looks well written, which is exactly the signal that makes people stop looking.
The other failure is reviewing everything. Review attention is the scarcest input in the whole system, and spending it on mechanical changes burns it. The workable shape is to review the diff for the parts you did not specify, and let the automated checks carry the parts that are mechanical.
Related terms
Automated Review
An agent reviewing another agent's work, often with a different model or system prompt. It forms a judgement.
Code Review
A second pass by someone who did not write the code, looking for what the author missed.
Human in the Loop
A person who reviews, approves, or corrects the agent's work at key points.
Over-reliance
Trusting the output because it is well written, not because it is right.
Vibe Coding
Writing software by describing what you want and accepting the result without reading it closely.
Want this working inside your team?
It starts with one conversation about your workflow.